Linearizing Equations Handout

Wilfrid Laurier University

Terry
c Sturtevant1 2

January 21, 2009

1 Physics Lab Supervisor

2 This document may be freely copied as long as this page is included.
ii

January 21, 2009

Chapter 1

Linearizing Equations

1.1 Theory
Often, the point of a scientific experiment is to try and find empirical val-
ues for one or more physical quantities, given measurements of some other
quantities and some mathematical relationship between them. For instance,
given a marble has a mass of 5 g, and a radius of 0.7 cm, the density of the
marble can be calculated given that v = 4/3πr3 and ρ = m/v. (For the sake
of simplicity, uncertainties will be ignored for now, although the calculation
of those should be familiar by now.)
Many times, however, rather than having one measurement of a quan-
tity, or set of quantities, we may have several measurements which should
all follow the same relationships, (such as if we had several marbles made
of the same material in the example above), and we wish to combine the
results. The usual way of combining results is to create a graph, and ex-
tract information (such as the density) from the slope and y–intercept of the
graph.
One may be tempted to ask why a graph should be better than merely
averaging all of the data points. The answer is that an average is completely
unbiased. The variation of any one point from the norm is no more or less
important than the variation of any other point. A graph, however, will show
any point which differs significantly from the general trend. Analysis of the
graphical data (such as with a least squares fit) will allow such “outliers” to
be given either more or less weight than the rest of the data as the researcher
deems appropriate. Depending on the situation, the researcher may wish to

January 21, 2009

2 Linearizing Equations

Figure 1.1: Non-linear equation

January 21, 2009

1.1 Theory 3

verify any odd point(s), or perhaps the trend will indicate that a linear model
is insufficient. In any case, it is this added interpretive value that a graph
has which makes it preferable.

A plot is better than an average since it may indicate systematic errors in

the data.
The value in fitting the data to an equation is that once the fit has been
done, rather than continuing to work with a large amount of data, we can
simply work with the parameters of our fit and their uncertainties. In the
case of a straight line, all of our data can be replaced by four quantities;
m, ∆m, b and ∆b.
A fit equation replaces a bunch of data with a few parameters.

The reason a linear graph is so useful is that it’s easier to identify whether
a line is straight than it is to identify whether it looks more like y = x2 or
y = x3 , for instance.
A straight line is easy to spot with the unaided eye.

If the data fits an equation of the form y = mx + b, then it is easy to plot

a straight line graph and interpret the slope and y–intercept, but it is rarely
that simple. In most cases, the equation must be modified or linearized
so that the variables plotted are different than the variables measured but
produce a straight line.
Linearizing equations is this process of modifying an equation to pro-
duce new variables which can be plotted to produce a straight line graph. In
many of your labs, this has been done already.
Look again at y = mx + b. Note that y and x are variables, (as each can
take on a range of values), while m and b are constants, (as there is only one
value for each for all of the data points). We can linearize an equation if we
can get it in the form

variable1 = constant1 × variable2 + constant2

There are a few things to note:

1. Several constants combined together produces another single constant.

2. Powers or functions of constants are also constants.

January 21, 2009

4 Linearizing Equations

3. Constants may have “special” values of 0 or 1 so they appear “invisi-

ble”. For example
y = mx
is still the equation of a straight line, where b = 0. As well,

y=b

is the equation of a line where m = 0.

4. Variables may be combined together to form new variables.

5. Powers or functions of variables are also variables.

Note that linearizing an equation will produce expressions for the slope and
y–intercept which depend only on the constants in the original equation, not
on the original x and y variables. This means that the constants can be
related to the slope and y–intercept rather than the original variables.

1.1.1 Techniques for Linearization

If a relationship involves only multiplication and division, (including powers),
then logarithms can be used to linearize. Sometimes taking roots or powers
of both sides of an equation will help.

1.1.2 Procedure for Linearization

The steps are as follows:

1. Rearrange the equation to get one variable (or a function of it) on the
left side of the equation; this becomes your y variable.

2. Regroup the right side of the equation to create a term containing the
other variable (or some function of it).

3. Use the left-side variable (or the function of it) as your x variable, and
then your slope should be whatever multiplies it; your y intercept is
whatever additive term is left over.

January 21, 2009

1.1 Theory 5

Note: It is important to realize that you don’t need to understand an equa-

tion to linearize it; all you have to know is which parameters are variables (ie.
things you have data for), and which parameters are constants (ie. things
you want to calculate). Of course different experiments involving the same
relationship may make different parameters variable, and so how an equa-
tion is linearized will depend on the data used. To again consider the above
example: The original equations were

v = (4/3)πr3 (1.1)

and
ρ = m/v (1.2)
where the quantities m and r are measured. (ie. We have several marbles of
the same material, so we can get several measurements of m and r, but we
expect ρ to be the same for all of them.) Thus for this situation, m and r
are variables, and ρ is a constant. We can combine the two equations to get
m
ρ= (1.3)
(4/3)πr3
or
3m
ρ= (1.4)
4πr3
This equation has a constant on one side, and a mixture of variables and
constants on the other. First we should rearrange it to get a variable on the
left hand side. Suppose we rearrange the equation, giving

m = (4/3)πρr3 (1.5)

This leaves a variable on the left. From this point on, there are two main
possibilities for how to proceed: 1

Method I
Now we can create a new variable, Y such that

Y =m
1
Usually the process is not as explicit as this. ie. one doesn’t usually create an X and
a Y , but doing this illustrates the procedure.

January 21, 2009

6 Linearizing Equations

By the rule about powers of variables being variables, then we can create a
new variable X given by
X = r3
Then equation 1.5 above becomes

Y = (4/3)πρX (1.6)

since π is a constant, and ρ should be, and using the rule that combinations
of constants produce constants, then we can define M , a constant, (not the
same as m), as
M = (4/3)πρ
so equation 1.6 becomes
Y = MX + 0
which is the equation of a straight line. (In the case, B, the y–intercept
is zero.)2 So if we plot our “modified” variables, we should get a straight
line, passing through the origin with a slope M . How can we get ρ from the
graph? Well, from above
M = (4/3)πρ
so
3M
ρ=
4π
where M is the slope of the graph.

Method II
We can take logarithms of both sides, so that Y such that equation 1.5 above
becomes
ln m = ln ((4/3)πρ) + ln r3 (1.7)
grouping the terms so one only contains constants (and so the combination
should be constant) and one only contains the variable r. We can bring down
the exponent so equation 1.7 becomes

ln m = ln ((4/3)πρ) + 3 ln r
2
Occasionally we can get a situation where the slope is similarly “invisible”, if it is 1
or 0.

January 21, 2009

1.1 Theory 7

r3

Figure 1.2: One linearization

January 21, 2009

8 Linearizing Equations

Now we can create new variables, Y such that

Y = ln m
and
X = ln r
which is the equation of a straight line. So if we plot our “modified” variables,
we should get a straight line. How can we get ρ from the graph? Well, from
above
B = ln ((4/3)πρ)
so
3 B
ρ= e
4π
where B is the y-intercept of the graph. (In this case, the value you get from
the graph for the slope should suggest whether the fit is a good one.)
Remember that after linearization, our results depend on our graphical quan-
tities of the slope and the y-intercept, rather than on the original measured
quantities.

1.1.3 Choosing a Particular Linearization

Often there may be more than one linear form for the equation so there may
be more than one “right answer”. In this case, there are a few things which
may help you choose.

Simple variables
A preferable linearization is one which most simplifies understanding the
graph or interpreting the results. For instance, in the above example, it
would have been possible to use (4/3)πr3 instead of r3 as our x variable, but
that would make confusing axis scales and/or units (although it would have
made the slope be ρ with no calculation).

Spread of data
The spread of data will be different for each linearization. A graph with
points which are more equally spaced is generally preferable to one where
the points are concentrated in one area.

January 21, 2009

1.1 Theory 9

ln (m)

ln (r)

Figure 1.3: Another linearization

January 21, 2009

10 Linearizing Equations

Size of error bars

Like the spread of data, the size of the error bars will be different for each lin-
earization. A graph with more equally sized error bars is generally preferable
to one where the error bars vary greatly in size for different points.
Usually it is preferable to separate variables and constants as much as possible
in your linearization so that graph variables are easily related to experimental
ones.

1.1.4 Uncertainties in Results

After determining how equation parameters relate to graphical quantities
as above, uncertainties can be determined as usual. In the above example
Method I gives
3∆M
∆ρ =
4π
while for Method II
3 B
∆ρ = e ∆B
4π
or
∆ρ = ρ∆B
For instance, from the example above, using Method I we would say:
• Plot m vs. r3 . (In other words, the independent (x) variable is r3 and
dependent (y) variable is m.)
• The uncertainty in the dependent variable is ∆m.
• The uncertainty in the independent variable is 3r2 ∆r.
• The slope of the graph will be M = (4/3)πρ.
• The y-intercept should be zero3 .
3M
• The density will be determined from the slope by the equation ρ = 4π
.
• The uncertainty in the density will be determined from the slope by
the equation ∆ρ = 3∆M
4π
.
3
If the y-intercept turns out to be something other than zero, then there is some
systematic error in our experiment.

January 21, 2009

1.2 Recap 11

1.2 Recap
By the end of this exercise, you should understand the following terms:

• linear graph

• linearized equation

January 21, 2009